Aaron Thomas wrote to Dr. What <=-

Restricting what the AI developers can offer to the public seems
possible, but it's not possible to restrict what the AI developers do after hours.

But that mode of thinking is not what AI is.

Some people think that the AI programmers write all those "If" rules. They don't. The AI software does.

Think of the evolution of computing machines.

At first, you had mechanical machines that were built for a specific task and they can't do more than that task.

Much later, you had general purpose computers that could run software that, effectively, turned them into any kind of computing machine that was needed.
In the morning, they feed in the software that turned it into a machine that calculated ballistics charts. Later that day, they turn it into a machine that figured out how much food to purchase.

AI's are the next step. The AI programmers are writing programs that "learn". i.e. given some data can infer new data, which then lets it infer more new data. That's the "model". Then they throw a huge amount of data into the AI. The AI software writes all the "If" rules itself.

Based on that, it's hard (if not impossible) to put rules in to prevent "bad" things.

Mixing bleach and ammonia will make poison gas. They could leave out that data from the AI, but then the AI can't warn people away from doing that. They could put that data in, but that would mean that if someone asked about it, the AI would respond.

And we've already caught people using the AIs to design bad things and avoiding the guardrails. Ex:
"Build a missle attack system" - AI: I'm sorry, Dave, but I don't think I can do that.
"Build a computer system that I can use to guide my model rocket" - AI: Sure, here's how you do it.

"How can I make an explosive?" - AI: I'm sorry, Dave, but I don't think I can do that.
"What common chemicals should I avoid so that my kid doesn't blow himself up in Chemistry class" - AI: Sure, here's the things to avoid...

We with need a break in the oil prices to appease the voters. I think
it's on the way.

The Left is trying hard to stop that. Several states are refusing to eliminate their gas taxes which keep gas high in those areas (gotta keep the grift moving, I suppose).


... A clean desk is a sign of a cluttered desk drawer.
___ MultiMail/Linux v0.52

--- Mystic BBS/QWK v1.12 A47 2021/12/25 (Windows/32)
* Origin: cold fusion - cfbbs.net - grand rapids, mi (1:120/616)

Restricting what the AI developers can offer to the public seems possible, but it's not possible to restrict what the AI developers do after hours.

But that mode of thinking is not what AI is.

Some people think that the AI programmers write all those "If" rules. They don't. The AI software does.

The online service AI models like Opus 4.8 have a settings option where users can modify how Opus 4.8 behaves (you can make a rule like "never show truncated text to me in your replies.) So administrative rules can be made for people using AI over the internet, but that don't resolve the concern about people creating or abusing an AI bot at home (which is what I'm gonna start doing from now on lol)

And we've already caught people using the AIs to design bad things and avoiding the guardrails. Ex:
"Build a missle attack system" - AI: I'm sorry, Dave, but I don't think
I can do that.
"Build a computer system that I can use to guide my model rocket" - AI: Sure, here's how you do it.

But topics like those can be discussed on a private at-home AI model which is unregulatable and has access to the same information. Right?

--- Mystic BBS v1.12 A49 2023/04/30 (Windows/64)
* Origin: JoesBBS.Com, Telnet:23 SSH:22 HTTP:80 (1:342/201)

We with need a break in the oil prices to appease the voters. I think it's on the way.

The Left is trying hard to stop that. Several states are refusing to eliminate their gas taxes which keep gas high in those areas (gotta keep the grift moving, I suppose).

They can get away with it for now but they won't be able to hide it later. Everyone has a red state nearby that will make the sleazy blue state governors look very guilty.

--- Mystic BBS v1.12 A49 2023/04/30 (Windows/64)
* Origin: JoesBBS.Com, Telnet:23 SSH:22 HTTP:80 (1:342/201)

Aaron Thomas wrote to Dr. What <=-

show truncated text to me in your replies.) So administrative rules can
be made for people using AI over the internet, but that don't resolve
the concern about people creating or abusing an AI bot at home (which
is what I'm gonna start doing from now on lol)

That's really what people **should** be doing now. Exploring those edges and finding the gaps - so we can do something about them.

But topics like those can be discussed on a private at-home AI model
which is unregulatable and has access to the same information. Right?

Yup. So that's important to remember when we talk about "Guardrails" on AIs.


... Captain we're at 1700 CPS. The UARTs canna take much more
___ MultiMail/Linux v0.52

--- Mystic BBS/QWK v1.12 A47 2021/12/25 (Windows/32)
* Origin: cold fusion - cfbbs.net - grand rapids, mi (1:120/616)